CR-IoTNet: Machine learning based joint spectrum sensing and allocation for cognitive radio enabled IoT cellular networks

Abstract

Abstract In recent years, the Internet of Things (IoT) paradigm has gained much popularity due to its potential ability to integrate the physical world with the digital world. However, this digital revolution is driving an insatiable demand for wireless spectrum, leading the upcoming novel personal, social and industrial IoT applications competing for limited spectrum resources. One of the main challenges that may hinder the vision of IoT is the scarcity of radio spectrum. Therefore, in order to cope with the challenge of spectrum scarcity brought about by the unprecedented number of IoT devices, this paper proposes to integrate cognitive radio (CR) technology with the IoT paradigm (CR-IoTNet). The CR is realized as a promising solution to improve the spectrum utilization and alleviate the spectrum scarcity in wireless networks. In CR, spectrum sensing is widely recognized as a key technology that enables secondary users (SUs) to detect spectrum holes and have the opportunity to access unoccupied spectrum. The CR-IoTNet, which is composed of multiple primary user (PU) base-stations and SU devices as IoT smart objects, performs the joint spectrum sensing and optimal allocation of spectrum to the requesting SU-IoT devices in the network through an intelligent fusion center (IFC). Furthermore, we employ support vector machines (SVM) in CR-IoTNet, enabling it to learn and adapt to the changing network dynamics, and identify the PU spectrum usage based on the established multi-class ( J × 6 ) -D feature set. The performance of CR-IoTNet is evaluated across several key factors, where simulation results validate the efficacy of the proposed framework in achieving high reliable identification, classification and allocation of unoccupied frequency bands in PU spectrum, especially in key areas of low signal-to-noise ratio (SNR). In addition, the trained SVM classifier obtained 95.11% accuracy, with 92.67% true positive rate and 96.34% true negative rate in CR-IoTNet, reflecting the optimal allocation of PU nodes for spectrum access.